Observational Quantification of Tropical High Cloud Changes and Feedbacks

The response of tropical high clouds to surface warming and their radiative feedbacks are uncertain. For example, it is uncertain whether their coverage will contract or expand in response to surface warming and whether such changes entail a stabilizing radiative feedback (iris feedback) or a neutral feedback. Global satellite observations with passive and active remote sensing capabilities over the last two decades can now be used to address such effects that were previously observationally limited. Using these observations, we show that the vertically averaged coverage exhibits no significant contraction or expansion. However, we find a reduction in coverage at the altitude where high clouds peak and are particularly radiatively‐relevant. This results in a negative longwave (LW) feedback and a positive shortwave (SW) feedback which cancel to yield a near‐zero high‐cloud amount feedback, providing observational evidence against an iris feedback. Next, we find that tropical high clouds have risen but have also warmed, leading to a positive, but small, high‐cloud altitude feedback dominated by the LW feedback. Finally, we find that high clouds have been thinning, leading to a near‐zero high‐cloud optical depth feedback from a cancellation between negative LW and positive SW feedbacks. Overall, high clouds lead the total tropical cloud feedback to be small due to the negative LW‐positive SW feedback cancellations.


Introduction
Earth system models project that warmer surface temperatures caused by rising atmospheric CO 2 concentrations will render clouds less effective at cooling the planet, causing further warming, that is, a net positive feedback (Zelinka et al., 2020).However, the precise magnitude of the cloud feedback remains unsettled due to uncertainties within individual components of the cloud feedback (Gettelman & Sherwood, 2016).Recent advances in research on low clouds using observations, process-resolving models, and climate models have translated to reduced uncertainty in low cloud feedbacks and hence climate sensitivity (Bretherton, 2015;Brient & Schneider, 2016;Cesana & Del Genio, 2021;Klein et al., 2017;Medeiros et al., 2008;Myers et al., 2021;Zelinka et al., 2020).Such an advance has not yet happened for high clouds, making it one of the most uncertain cloud feedbacks.For example, the tropical high cloud cover (HCC) feedback is the most negative cloud feedback and has the greatest uncertainty according to an expert assessment based on synthesizing the literature (World Climate Research Program (WCRP) expert assessment; Sherwood et al., 2020).High cloud feedbacks remain uncertain due to the underlying uncertainty in how their properties-amount, altitude, and optical depth-respond to warming.
The tropical HCC feedback has been controversial for some time (Hartmann & Michelsen, 2002;Lin et al., 2002;Lindzen et al., 2001;Ramanathan & Collins, 1991).Some studies argue that it constitutes a strong stabilizing negative feedback for the climate system, termed the "iris feedback" (Lindzen et al., 2001), while others argue that it has a neutral feedback overall (Mauritsen & Stevens, 2015).The WCRP's assessed negative feedback was based on a single observational study (Williams & Pierrehumbert, 2017).Thus, there is a significant need for more observational studies to better understand this radiative feedback.This tropical HCC radiative feedback is determined by whether or not high clouds increase or decrease in horizontal coverage in response to surface warming.Furthermore, the feedback depends which types of clouds increase or decrease, thinner types that have a net warming effect on the TOA budget, or thicker types that have a net cooling effect (Hartmann et al., 1992).Lindzen et al. (2001)'s iris feedback arose from a significant decrease in HCC.Hereafter, we refer to an HCC decrease as an iris effect (see Column Iris and Peak Iris below) and the radiative impact is termed the iris feedback (see Iris Feedback below).Our study will provide a much-needed observational quantification of both.
The varying definitions of "iris" can cause confusion, so we classify the three different iris phenomena investigated in this study.
• Iris Feedback: A radiative feedback in response to a decrease in high cloud cover (Lindzen et al., 2001).
• Column Iris: A net decrease in coverage in high clouds in the vertical column (Lindzen et al., 2001).
• Peak Iris: A decrease in high cloud cover at the peak coverage level (Bony et al., 2016;Saint-Lu et al., 2020).
While there is much uncertainty about how warming may change the horizontal distribution of high clouds, vertical changes are robust in Earth system models (Zelinka & Hartmann, 2010): high clouds rise with warming.This is because of the Fixed Anvil Temperature (FAT) hypothesis (Hartmann & Larson, 2002).The tropics are approximately in radiative-convective equilibrium, and clear-sky radiative cooling decreases rapidly around 13 km/the 215 K isotherm due to fundamental thermodynamics and radiative transfer of H 2 O. Consequently, convective heating decreases, too.Combined with mass continuity constraints, the FAT hypothesis postulates that detrainment from deep convection will occur at this 215 K isotherm in the tropics.The temperature at detrainment level is nearly invariant with global warming because H 2 O is constrained by the Clausius-Clapeyron relation.Since the upper troposphere also gets much warmer with global warming, for high clouds to maintain the same temperature, they must rise.If high clouds rise isothermally while the surface warms, then the radiative effect from these clouds increases: a positive cloud feedback (e.g., Stauffer & Wing, 2023;Zelinka & Hartmann, 2010).This hypothesis was modified later to account for the slight warming of clouds due to upper-tropospheric stability increases and is known as the Proportionately Higher Anvil Temperature (PHAT) hypothesis (Zelinka & Hartmann, 2010).
The FAT hypothesis has conflicting support in cloud-resolving models (Kuang & Hartmann, 2007;Seeley et al., 2019) and has not been demonstrated in decadal trends in vertically-resolved cloud cover observations.Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO) fills this gap by providing an unprecedented long-term, continuous, vertically-resolved, high-resolution, global record of cloud cover from 2006 to 2020 (Chepfer et al., 2010).Furthermore, the CALIPSO mission ended on 1 August 2023, and the investigation of HCC trends over the complete CALIPSO period available is yet to be conducted.We will use these CALIPSO observations to show that high clouds are indeed rising and in conjunction with other satellite observations we will quantify the radiative impact of these high cloud altitude changes.
This study thus has twin aims: (a) quantify the changes in high cloud properties such as amount, altitude, and optical depth using both active and passive remote sensing and (b) quantify the resultant high cloud radiative feedbacks using passive remote sensing.In the next section, we first describe the satellite observations used to better understand the behavior of vertically-resolved high cloud cover changes (Section 2.1) and then we describe the standard methodology used for cloud feedback decompositions in climate models that we now apply to observations (Section 2.2).Following the methodological section, we present our results on changes in cloud properties (Section 3.1) and their accompanying radiative feedbacks (Section 3.2).Finally, we close with the conclusions of this study, a comparison of our observational feedbacks with the WCRP assessment, and future directions (Section 4).

Data
Our study utilized active (lidar) remote sensing, passive remote sensing, surface measurements, and reanalyzes products.We used lidar measurments from the GCM-Oriented CALIPSO Cloud Product (GOCCP) (Chepfer et al., 2010).CALIPSO GOCCP derives its data from the Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP) Level 1 data at native resolutions.CALIPSO GOCCP provides monthly-mean vertically-resolved profiles of cloud cover spanning June 2006-December 2020 at each 2°× 2°gridbox across the planet.It has 40 vertical levels from sea level to 19.2 km with 480 m resolution.We refer to tropical (30°S-30°N) high clouds as the vertically-resolved cloud cover profile above 6.5 km.Since most of the previous studies used only passive space sensors that could not detect very thin clouds, CALIPSO cloud retrievals, which are particularly sensitive to high clouds and robust to high cloud overlap considerations (Chepfer et al., 2010), will yield additional insights into tropical high cloud changes.
We note that thick high clouds can cause difficulty in detecting low clouds by attenuating the lidar signal, however, as our study focuses on high clouds (z > 6.5 km), lidar attenuation should not bias our high cloud results (Chepfer et al., 2010).Furthermore, CALIPSO GOCCP provides monthly-mean and gridded cloud fields, and we further compute tropical-averages for our analysis, so the temporal and spatial averaging would further reduce any potential ability of attenuation or overlap considerations to affect our study, which is already likely negligible for high clouds as discussed above (Chepfer et al., 2010).We note that because the South Atlantic Anomaly (SAA) affects the CALIPSO lidar instrument's retrievals (Noel et al., 2014), we masked the region of 29°S-11°S, 85°W-15°W in the tropics and 49°S-1°S, 99°W-1°W in the globe to prevent spurious trends from affecting the underlying signals.Earth's magnetic field has a weak spot over the SAA, which allows more ionized particles to reach Earth's upper atmosphere and consequently affect spaceborne electronics and lidar-based retrievals over the SAA (Noel et al., 2014;Figure S1 in Supporting Information S1).Results are insensitive to different definitions of the SAA (Section 3.1).We also used CALIPSO's breakdown of cloud cover into optically thick (τ > 3 5; where τ is the optical depth) and optically thin (τ < 3 5) clouds.
We used passive remote sensing measurements from the Clouds and Earth's Radiant Energy System (CERES) FluxByCldTyp (i.e., Flux By Cloud Type, FBCT) data product (Sun et al., 2022).FBCT provides cloud properties and top-of-atmosphere (TOA) radiative fluxes binned by pressure and optical depth (Sun et al., 2022).There are six optical depth bins and 7 cloud effective pressure bins, together creating a joint histogram of cloud types.Horizontal resolution is 1°× 1°and temporal resolution is monthly with data during July 2002-December 2020 being used.FBCT measurements provide data for non-overlapping cloud layers (Sun et al., 2022) and although low-cloud retrievals may be biased due to overlying thin cirrus clouds (Loeb et al., 2003), our study focuses on high clouds which are less affected by such biases (Kim & Ramanathan, 2008).The cloud type fluxes are the contribution of each cloud type to the TOA flux, that is, the sum of fluxes across cloud types amounts to the TOA flux for any given location and month.The FBCT radiation quantities used are clear-sky fluxes and overcast fluxes per cloud type which are then used for calculating the radiative kernels and cloud feedbacks.
The FBCT cloud properties used are cloud cover (cc), liquid cloud cover, ice cloud cover, liquid water path (LWP), and ice water path (IWP).We weight the cloud type LWP and IWP by the liquid cloud fraction and ice cloud fraction, respectively.The weighted sum of the IWP (or LWP) cloud types amount to the total IWP (or LWP) for a given location and month.The zonal-mean pattern of climatological IWP is consistent with the Intergovernmental Panel on Climate Change Fifth Assessment Report (Boucher et al., 2013;Figure S2 in Supporting Information S1).The monthly-mean surface temperature reconstruction from NASA GISTEMP (Lenssen et al., 2019) was used for calculating the sensitivity of cloud property changes to surface temperature as well as radiative feedbacks.
Finally, we used two reanalysis products to supplement our observational analysis: European Centre for Medium-Range Weather Forecasts Reanalysis 5 (ERA5; Hersbach et al., 2020) and National Centers for Environmental Prediction (NCEP) reanalysis (Saha et al., 2010).We used ERA5 temperature and cloud cover profiles to diagnose the warming of high clouds and the upper troposphere.We used NCEP tropopause data to display the climatological tropopause height across latitudes (Hoffmann & Spang, 2022).

Methods
Climatologies were calculated as time-averages during June 2006-December 2020 for CALIPSO and July 2002-December 2020 for FBCT.Monthly anomalies were calculated by removing the monthly climatology from the data.Sensitivities were calculated by regressing the anomalies in a tropical-mean cloud property against anomalies in tropical-mean surface temperature (TMST) at each vertical level or cloud type.Although Section 3.1 focuses on regressions against anomalies in TMST, temporal trends were also calculated in CALIPSO over the complete period.The trend was calculated as the linear fit through the time series of anomalies in cloud cover at each vertical level.Using the standard error (S.E.) associated with the linear fit, a 95% confidence interval (CI; 1.96 × S.E.), was attached as an estimate of uncertainty.
Cloud radiative kernels (K), the sensitivity of TOA radiative fluxes to cloud cover changes, and cloud feedbacks (λ cloud ), the sensitivity of cloud-induced radiative anomalies to surface temperature changes, are calculated following the Zelinka et al. (2012aZelinka et al. ( , 2012bZelinka et al. ( , 2013) ) methodology that is widely used for quantifying cloud feedbacks (Section 3.2).First, K is calculated by: where t represents time, lat represents latitude, τ represents optical depth, p represents pressure level, R clr represents clear-sky radiative flux, and R ovc represents the overcast-sky radiative flux for each cloud type.The latter two quantities in FBCT observations are averaged over all longitudes (zonal-mean) to reduce the effect of missing data in observations.The numerator in Equation 1 represents the overcast-sky cloud radiative effect for each cloud type, latitude, and month (Hartmann et al., 2001;Kubar et al., 2007).Dividing it by 100 provides the sensitivity of TOA radiation to a change in cloud cover, that is, the kernel, in units of Wm 2 % 1 .The kernel is computed for longwave (LW) and shortwave (SW) fluxes and the sum of the two produces the "Net" kernel.
The cloud-induced radiative anomaly (ΔR) is obtained by multiplying the kernel (Equation 1) by the cloud cover anomalies (Δcc) time series: Using cloud cover that spans longitudes does not change the results.We area-average ΔR to obtain the tropicalaverage flux: ΔR(t,τ,p).The tropical λ cloud is then obtained by the slope of the linear regression of ΔR against global-mean surface temperature (GMST) anomalies (ΔT s ): λ cloud can be decomposed into low (p > 680 hPa or z < 3.2 km), middle (440 hPa < p < 680 hPa or 3.2 km < z < 6.5 km), and high (p < 440 hPa or z > 6.5 km) cloud feedbacks based on CALIPSO GOCCP and the International Satellite Cloud Climatology Project (ISCCP) classification (Chepfer et al., 2010;Hartmann et al., 1992;Rossow & Schiffer, 1991): For example, the high cloud feedback (λ clh ) is: Finally, λ cloud can be further decomposed into amount, altitude, optical depth (τ), and residual feedbacks: Journal of Geophysical Research: Atmospheres λ clh is similarly decomposed.We refer the reader to Zelinka et al. (2013)'s Appendix B for details on calculating the terms on the right-hand side of Equation 6.

Results
In Section 3.1, tropical climatologies and changes in cloud cover/amount are investigated in active (CALIPSO) and passive (FBCT) satellite measurements.Next, tropical and zonal-mean climatologies and changes are investigated for cloud altitude and temperature in active satellite measurements (CALIPSO) and present-day reanalysis data (ERA5).Finally, tropical climatologies and changes for cloud optical depth properties (IWP) are investigated in passive satellite measurements (FBCT).In Section 3.2, tropical cloud radiative kernels are first constructed to investigate the sensitivity of the radiation budget to changes in clouds for different cloud types using passive satellite measurements (FBCT).Next, tropical cloud feedbacks are computed per cloud type using passive satellite measurements (FBCT).Finally, tropical high cloud feedbacks are decomposed into amount, altitude, optical depth, and residual components using passive satellite measurements (FBCT).

Cloud Property Changes
The CALIPSO-observed, tropical-average, climatological cloud cover profile shows a peak in HCC at approximately 13 km, for reasons that are explained in Section 1 (Figure 1a). Figure 1b shows how this profile responds to surface temperature anomalies.When the surface warms, a dipole emerges with cloud cover below ≈13 km decreasing and increasing above.When examining the entire column above 6.5 km, there is no significant change in the vertical-average: 0.13 ± 0.16%/K (not significant with 95% CI), that is, there is no Column Iris effect (Section 1).These results are insensitive to the surface-temperature dataset used (Figure S3 in Supporting Information S1).We note that the peak cloud coverage at 13 km (marked by dots in Figures 1a and 1b), which is radiatively-relevant for feedbacks, shows a decrease of 0.29 ± 0.28%/K in response to TMST anomalies (significant with 95% CI), in line with the response of peak high cloud cover to warming seen in other studies (Bony et al., 2016;Saint-Lu et al., 2020).FBCT shows a peak coverage at 310-180 hPa (Figure 1c) and also shows the dipole effect but the dipole differs for thin and thick clouds: for thin clouds (τ ≤ 3.55) the decrease in cloud is at 440-310 hPa with an increase above, and for thick clouds (τ > 3.55) cloud decreases below 180 hPa and shows slight increases above (Figure 1d).We note that the vertical dipole may appear at a lower altitude for thinner clouds because the passive instruments place the thin clouds too low in the atmosphere in the presence of multi-layered clouds (Mitra et al., 2021).The decreases in optically thick clouds in FBCT (310-180 hPa) are radiatively-relevant and will be discussed in Section 3.2.
Figure 2a shows the temporal trend of the tropical-average cloud cover at each vertical level.We found that high cloud cover decreases below the peak coverage and increases above it, that is, a dipole of strong decrease-increase that mirrors the cloud cover sensitivity of Figure 1b.These trend results are significant at each level above 8 km (95% CI) and are insensitive to using a shorter time-period (2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015), monthly-mean versus annual-mean data, or a wider definition for the SAA (Figure S4 in Supporting Information S1).Vertically averaged above 6.5 km, the net trend is not significant: 0.07 ± 0.08%/decade (95% CI).Thus, when considering the cloud cover in the tropical upper troposphere, the CALIPSO observations again show no evidence of a Column Iris effect (Section 1).A simpler way to visualize these changes is to compare HCC at the beginning and end of the time-period as in Zelinka and Hartmann (2010).The end of the period (2018-2020) has clouds higher than earlier in the period (2007)(2008)(2009) and subtracting these two periods yields a dipole (Figure S5 in Supporting Information S1).The dipoles shown in Figures 1b and 2a are thus a vertical shift of clouds.The trend dipole occurs in both opaque and non-opaque clouds (Figure S6 in Supporting Information S1), and therefore this dipole is a robust result for thick and thin high clouds.The dominant signal in the lidar-satellite cloud record is thus the rise of clouds, not any net horizontal changes, that is, no Column Iris effect (Section 1).
In Figures 2b and 2c we extended our analysis to the rest of the planet.Figure 2b's latitude-height plot shows that high cloud coverage is largest in the deep tropics and the maximum cloud top height decreases poleward.This poleward decrease in maximum cloud top height is consistent with the tropopause height's equator-topole characteristics as seen in Figure 2b (Gettelman et al., 2011).Figure 2c then shows the trend in the zonal-mean cloud cover profile.It shows that the high cloud coverage shifted upward across the globe, that is, the rise of high clouds extended to outside of the tropics.The differing altitudes of the tropical and extratropical dipoles cause a global dipole as well (Figure S7 in Supporting Information S1).This zonal-mean pattern is predicted by climate models (Sherwood et al., 2020;Thompson et al., 2019) but is now evident as trends in the lidar satellite record.These results match the average pattern seen in CMIP models (e.g., Figure 5a of Sherwood et al., 2020).Next, we evaluate whether these rising clouds are staying at a fixed temperature, in accordance with FAT, or warming, in accordance in with PHAT.Following Zelinka and Hartmann (2010, their Equation 5), we calculate high-cloud temperatures by:  where f = cc/100 and T are the cloud fraction (from CALIPSO GOCCP and ERA5) and atmospheric temperature (ERA5 only) at each vertical level, respectively.The two formulas are very similar and only differ in tropicalaveraging after cloud fraction and temperature are multiplied (Equation 7a) or tropical-averaging before cloud fraction and temperature are multiplied (Equation 7b).The sensitivity of the anomalies in these high-cloud temperatures to anomalies in TMST is listed in Table 1.We found that high clouds are warming at a rate of 0.58-0.86K/K, depending on the dataset and method used.For comparison, the atmosphere at 200 hPa is warming at a higher rate of 1.68 ± 0.17 K/K.As the warming rate of high clouds is lower than the upper-troposphere warming rate, tropical high clouds rise and warm, in-line with the PHAT hypothesis.
Finally, we investigate changes in optical properties of clouds.Climatological ice water path is limited to high clouds (P < 440 hPa; Figure 3a).With warming, IWP decreases for most cloud-types and summed together decreases 3.32 ± 3.78 gm 2 K 1 (not significant with 95% CI) (Figure 3b).In summary, high clouds are rising, warming, thinning, and when vertically averaged, are not contracting.Next, we quantify the radiative impact of these cloud property changes.

Cloud Feedbacks
To calculate the radiative impact of the cloud property changes in response to observed surface temperature changes shown in Section 3.1, that is, calculate short-term observed cloud feedbacks, we first compute the cloud radiative kernels in FBCT observations.Figure 4 shows these tropical-mean LW, SW, and Net cloud radiative kernels.These observational kernels agree with previous kernels (Sun et al., 2022;Zhou et al., 2013).Increases in cloud cover increase LW trapping (greenhouse effect) for all cloud types (Figure 4a).Increase in cloud cover at high altitudes are particularly effective at preventing LW radiation from escaping to space because the high clouds emit infrared radiation from a colder temperature compared to the clear-sky atmosphere.Clouds increase SW reflection (albedo effect) for all cloud types (Figure 4b).Increases in cloud cover at large optical depths and all altitudes are particularly effective at increasing reflection of sunlight.Large cancellations in the LW and SW kernels result in a Net kernel heating impact in upper thin clouds and cooling impact on optically thick low clouds (see Figure S8 in Supporting Information S1 for sum across all optical depths).Now that the radiatively-relevant clouds are known through these FBCT kernels, multiplying the kernel by the FBCT cloud cover change (Figure 1d) at each latitude and month will produce the FBCT-observed short-term cloud feedback.
Tropical LW cloud feedback is found to be negative, that is, diminishing warming, for most cloud types and when summed over all τ-p bins it amounts to a stabilizing feedback of λ LW cloud = 0.57 ± 0.34 Wm 2 K 1 (Figures 5a and   Journal of Geophysical Research: Atmospheres 10.1029/2023JD039364 6).This is largely from high clouds (λ LW clh = 0.39 ± 0.30 Wm 2 K 1 ), supplemented by low-mid clouds ( 0.18 Wm 2 K 1 ; combined low and mid-level clouds).There is a noticeable dipole effect of negative-positive cloud feedback in thin clouds at 440-180 hPa and thick clouds at 310-310 hPa, consistent with the rise of high clouds discussed in Section 3.1 and Zelinka and Hartmann (2011).Tropical SW cloud feedback is found to be positive, that is, amplifying warming, for most cloud types and when summed over all τ-p bins it amounts to a destabilizing feedback of λ SW cloud = 0.80 ± 0.52 Wm 2 K 1 (Figures 5b and 6).This is again largely from high clouds (λ SW clh = 0.48 ± 0.30 Wm 2 K 1 ), supplemented by low-mid clouds (0.32 Wm 2 K 1 ; combined low and mid-level clouds).The tropical net cloud feedback is found to be positive for most cloud types, but not statistically significant when summed over all τ-p bins (λ cloud = 0.23 ± 0.40 Wm 2 K 1 ), that is, it may amplify or diminish warming (Figures 5c and 6).This is because of large cancellations between LW and SW feedbacks in high clouds, leading to a weakly positive net high cloud feedback of λ clh = 0.09 ± 0.12 Wm 2 K 1 , supplemented by low-mid clouds (0.14 Wm 2 K 1 ).Additionally, we note that the net feedback for very thin cirrus (0.02-1.27 optical depth, 440-10 hPa) is negligible ( 0.01 Wm 2 K 1 ).
The FBCT-derived feedbacks are insensitive to using a time-varying or a time-averaged cloud radiative kernel (Figure 6).The FBCT total LW, SW, Net cloud feedbacks are in excellent agreement with Raghuraman et al. (2023aRaghuraman et al. ( , 2023b)'s benchmark cloud feedback estimates that use partial radiation perturbation experiments to remove cloud masking and effective radiative forcing from CERES Energy Balanced And Filled (EBAF) cloud radiative effect (CRE) satellite observational data (Figure 6).Given the close agreement with previous estimates, we next decompose the high cloud feedback into its contributions from amount, altitude, and optical depth changes with a residual.
The tropical high-cloud amount feedback is near-zero (λ clh,amt = 0.05 ± 0.06 Wm 2 K 1 , not significant (p > 0.05)), arising from strong cancellations between the LW and SW feedbacks (Figure 7).This is due to decreases in the radiatively-relevant cloud cover in Figure 1.Thus, there is an infrared Iris Feedback (Lindzen et al., 2001), however, as seen in Figure 7, the solar component has an equal and opposite reaction that cancels the infrared, leaving a negligible net feedback that has a near-zero impact on the heat budget.The tropical high-cloud altitude feedback is positive (λ clh,alt = 0.12 ± 0.02 Wm 2 K 1 , significant (p < 0.05)) arising from λ LW clh,alt (Figure 7).This  behavior can be explained by the fact that high clouds are rising (Figure 2) and are warming less than the upper troposphere (Table 1), yielding a small positive feedback.The tropical high-cloud optical depth feedback is nearzero (λ clh,τ = 0.06 ± 0.09 Wm 2 K 1 , not significant (p > 0.05); Figure 7).Decreases in IWP with warming (Figure 3) cause a negative λ LW clh,τ as more outgoing longwave radiation can pass through the clouds and a positive λ SW clh,τ because the albedo is reduced; these LW and SW effects nearly cancel each other.
The residual (λ res ) is because of covariances, for example, simultaneous changes in coverage, altitude, and optical depth.We note that replacing high clouds with "non-low" clouds, that is, including mid-level clouds (as in Zelinka et al., 2016), does not alter the magnitude of the λ clh,amt and λ clh,alt feedbacks but does increase the λ clh,τ feedback.This is due to including mid-level clouds that show a decrease in LWP (Figure S9 in Supporting Information S1) which causes a positive SW cloud optical depth feedback.
In summary, the tropical LW high cloud feedback (λ LW clh ) is negative because of the negative amount (λ LW clh,amt ) and optical depth (λ LW clh,τ ) components overwhelming the positive altitude component (λ LW  clh,alt ).This helps explain why the tropical LW cloud feedback (λ LW cloud ) is negative (Figure 6).The tropical SW high cloud feedback (λ SW clh ) is positive because of positive contributions from the amount (λ SW clh,amt ) and optical depth (λ SW clh,τ ) components.This helps explain why the tropical SW cloud feedback (λ SW cloud ) is positive (Figure 6).The short-term tropical high cloud feedback (λ clh ) and net cloud feedback (λ cloud ) are therefore small with large uncertainties as a result of these LW and SW cancellations (Figure 6).

Conclusions and Discussion
By combining CALIPSO observations of vertically-resolved cloud cover and CERES-FBCT radiation measurements, this study arrives at three principal conclusions.First, on interannual timescales, tropical high clouds display no net contraction of horizontal coverage when vertically averaged and a near-zero high-cloud amount feedback, disputing the Column Iris and strongly negative Iris Feedback hypotheses (Section 1).Second, tropical high clouds are rising but are also warming, leading to a positive but small high-cloud altitude feedback.Third, tropical high clouds are thinning, leading to a near-zero high-cloud optical depth feedback.These conclusions are discussed further below.
Our study analyzed the entire vertical structure of high clouds and not just the high cloud cover response at the peak level.Thus, the apparent discrepancy with previous studies that reported decreases in anvil cloud coverage (Peak Iris; Saint-Lu et al., 2020) are not necessarily inconsistent with our results of a neutral change in high cloud Journal of Geophysical Research: Atmospheres 10.1029/2023JD039364 cover (no Column Iris).The distinction is that by focusing on anvil cloud cover, previous studies were restricted to the negative component of the tropical dipole (Figure 1b, Figure S10 in Supporting Information S1).Changes at any particular altitude, when considered in isolation, may not be representative of the full high cloud profile that shows decreases as well as increases in high cloud cover.
Using CALIPSO satellite observations, we showed that tropical high clouds are rising.This builds on several recent studies that also found a rise in tropical high clouds using passive and active sensors as well as studies that projected their upward shift associated with warming (Aerenson et al., 2022;Chepfer et al., 2014;Höjgård-Olsen et al., 2022;Norris et al., 2016;Richardson et al., 2022;Saint-Lu et al., 2020;Takahashi et al., 2019).We show that this surface warming-induced rise elicits a vertical dipole in cloud cover, consistent with the FAT mechanism.These tropical high clouds, however, are also found to be warming, making the rise and warming of high clouds found here more consistent with the PHAT mechanism.It is worth noting that increasing static stability in the upper troposphere not only causes a warming but also causes a decrease in peak cloud coverage, making the PHAT and the stability iris hypotheses two sides of the same coin (also see Zelinka & Hartmann, 2010).
The WCRP assessment of climate sensitivity crucially depended on assessing cloud feedbacks and two of the six cloud feedback components were related to high clouds (Sherwood et al., 2020).The high-cloud altitude feedback was assessed to be 0.20 Wm 2 K 1 while we found this feedback to be smaller at 0.09 Wm 2 K 1 .However, we note that comparing our study with their assessed altitude feedback requires the prerequisite assumption that the response of high-cloud altitude to warming is the same in short-term (the observational record shown here) and long-term time scales.The tropical anvil cloud area feedback was assessed to be 0.20 Wm 2 K 1 , while we found this feedback to be much less stabilizing at 0.03 Wm 2 K 1 (tropical value scaled by 50% to obtain global-mean).Although the sum of these (0.06 Wm 2 K 1 ) ends up being close to their zero high-cloud feedback, they arrived at this via canceling amount and altitude feedbacks while we show that this arises from a sum of weak feedbacks.Myers et al. (2021)'s update to the WCRP assessment placed the total cloud feedback at 0.27 Wm 2 K 1 and our update with new high cloud constraints would increase this to 0.33 Wm 2 K 1 .If all else is equal, this would translate to a 0.1 K increase in climate sensitivity from 3 to 3.1 K.A cloud controlling factor analysis (Myers et al., 2021) for high clouds, could further constrain our findings on climate sensitivity.
The aforementioned tropical anvil cloud area feedback in the WCRP assessment was largely based on the study of Williams and Pierrehumbert (2017) (hereafter WP17) which was also based on interannual observations, so it is worthwhile to study the potential causes of the discrepancy between their study and ours.Using FBCT, we confirm their finding of a negative net cloud feedback over deep convective regions and that high clouds largely cause this.However, over the much larger tropical domain (30S-30 N), which WP17 did not consider, the high cloud feedback becomes positive.Therefore, not accounting for high clouds over the entire tropical domain likely causes the discrepancy and high clouds outside deep convective regions play an important role (Kubar et al., 2007).When calculating the tropical-wide net high cloud feedback over the shorter period of WP17, we find a near-identical tropical high cloud feedback (λ clh = 0.11 Wm 2 K 1 ) compared to the full period used in this study (λ clh = 0.09 Wm 2 K 1 ).Thus, the tropical high cloud feedback is more sensitive to spatial sampling than to temporal variations.
The tropical high-cloud amount and optical depth feedback results require further investigation in the context of long-term warming.The net high-cloud amount feedback (λ clh,amt ) is near-zero due to K Net clh being near-zero when summed across optical depths (LW-SW cancellations; Figure 4; Figure S8 in Supporting Information S1) and radiatively-relevant Δcc clh decreasing (Equation 2).Understanding why K Net clh is near-zero, that is, a balanced radiative effect, in the present-day climate and whether it will remain in a future warm climate are crucial to determining the high-cloud amount feedback (Gasparini et al., 2019;Hartmann & Berry, 2017;Wall & Hartmann, 2018).Understanding whether Δcc clh evolves differently in response to long-term warming is critical as the resulting high-cloud amount feedback could be different.Finally, the mechanism underlying a thinning of high clouds is unclear, for example, whether it is induced by precipitation efficiency changes (Ito & Masunaga, 2022;Lindzen et al., 2001;Lutsko & Cronin, 2018;Rapp et al., 2005) and to what extent lowfrequency variability (such as ENSO) versus secular warming trends contribute to the thinning should be addressed in future work.
Finally, we note that the observed cloud property changes have imprints of radiative forcing (aerosol indirect effect, greenhouse gas rapid cloud adjustments, etc.) that should be disentangled from warming-induced changes Journal of Geophysical Research: Atmospheres 10.1029/2023JD039364 in future work (Raghuraman et al., 2023a).Our observational decomposition, spanning a time period of nearly two decades, can now be used as a benchmark for climate models.Future work should conduct a hierarchy of simulations, ranging from long-term warming to short-term warming, to better understand cloud feedback by probing the sensitivities in the amount, altitude, and optical depth components.

Figure 2 .
Figure 2. Cloud property climatology and changes: altitude.(a) CALIPSO GOCCP trend in tropical-average cloud-cover profile above 6.5 km.Shading represent 95% CI.(b) CALIPSO GOCCP zonal-mean and time-mean cloud cover profile.(c) CALIPSO GOCCP zonal-mean cloud cover trends profile.In panels (b) and (c) the black line depicts the climatological tropopause height.Accompanying shading represents the minimum and maximum tropopause height during the year.Time period = June 2006-December 2020.

Figure 3 .
Figure 3. Tropical cloud property climatology and changes: optical depth.(a) CERES-FBCT ice cloud-fraction weighted ice water path (IWP) profile climatology.(b) CERES-FBCT ice cloud-fraction weighted IWP profile sensitivity to TMST.Black line represents the boundary for high clouds.
Note.ERA5's sensitivity of 200 hPa atmospheric temperature to anomalies in TMST is also given to depict magnitude of upper-tropospheric amplified warming.CALIPSO GOCCP and ERA5 high cloud temperature anomalies are regressed against GISTEMP and ERA5 anomalies in TMST, respectively.Error bars represent 95% CI.TMST = Tropical-Mean Surface Temperature.Time period = June 2006-December 2020.All estimates exclude SAA region.T clh1 and T clh2 given by Equations 7a and 7b, respectively.

Figure 4 .
Figure 4. Tropical-average LW, SW, and Net kernels calculated directly from CERES-FBCT observations (July 2002-December 2020).Black line represents the boundary for high clouds.

Figure 6 .
Figure 6.Tropical-average cloud feedbacks comparison.Filled bars represent significant feedbacks ( p < 0.05) while unfilled bars represent nonsignficant feedbacks ( p > 0.05).Benchmark (R23 PRP) refers to the cloud feedback derived from partial radiation perturbation (PRP) experiments inRaghuraman et al. (2023aRaghuraman et al. ( , 2023b) that remove cloud-masking and effective radiative forcing from the CERES Energy Balanced And Filled (EBAF) cloud radiative effect satellite observational data.

Figure 5 .
Figure 5. Tropical-average LW, SW, and Net cloud feedbacks calculated directly from CERES-FBCT observations (July 2002-December 2020).Black line represents the boundary for high clouds.

Figure 7 .
Figure 7. Tropical-average LW, SW, and Net high cloud feedbacks decomposed into their contributions from amount, altitude, optical depth changes, and a residual.Filled bars represent significant feedbacks ( p < 0.05) while unfilled bars represent non-signficant feedbacks ( p > 0.05).Error bars represent 95% CI.